Circular RNAs as novel biomarkers with regulatory potency in human diseases

Circular RNAs (circRNAs) are a large class of noncoding RNAs characterized with closed loop structures without 3′ and 5′ polar ends. They can roughly be divided into exonic circRNAs, exon–intron circRNAs and circular intronic RNAs. CircRNAs are characterized with stability, prevalence, specificity and conservation, which arouse great interest in circRNAs as disease biomarkers. Their abilities to sponge to miRNAs, cis-regulate parent genes, bind to proteins and encode proteins endow circRNAs a critical role of regulation in eukaryotic cells. This concise review focuses on circRNAs as functional biomarkers and therapeutic targets in both tumor and nontumorous diseases.

observed in the head cannot be detected in other tissues [21]. Tissue-specific expression is also observable in plants.
For instance, in polyploidy Gossypium species, the overall expression level of circRNAs is higher in ovule samples than leaf samples. In addition, more than 80% circRNAs express only in ovule tissues [22]. Moreover, circRNAs are expressed in a developmental-stage specific manner. In mammalian and Drosophila brains, the global expression of circRNAs varies in different stages. For instance, mouse circRNAs derived from the Staufen2 gene exhibit reciprocal expression during neuronal differentiation [9]. In Drosophila, circRNA accumulation is associated with the aging process of the brain. The level of brain circRNAs is elevated from embryo to larva and pupa, and is even higher in adult heads [21].

Conservation
A small proportion of circRNAs is highly conserved across different species. In a research performed by Dong et al., the authors found about 15,000 circRNAs in both human and mouse genomes, indicating that 15% of total human circRNAs and 40% of total mouse circRNAs are conserved [23]. In another study concerning circRNA expression in heart, the authors discovered that about 10% cardiac circRNAs are conserved across human, mouse and rat, and about 30% cardiac circRNAs can be found simultaneously in mouse and rat [24]. In a recent study conducted by Stoll et al., among the 3441 explored human pancreatic islet circRNAs, 497 orthologous circRNAs can be determined in parallel mouse islet samples [25]. The conserved expression of circRNAs is associated with complementary intronic sequences flanking back-spliced exons [23]. Orthologous circRNAs across species often exhibit longer flanking introns than species-specific ones [22].

Nomenclature
With the advancement of next generation sequencing and bioinformatic technology, a large number of circRNAs are detected for the first time. Some newly discovered circRNAs are nominated by their founders concerning the type, location or function of circRNAs ( Figure 1). For example, ci-ankrd52, a ciRNA generated from parent gene ANKRD52, is named after its genome location. The name of a well-known ecircRNA, ciRS-7, contains the meaning of circRNA sponge for miR-7. However, such nomenclature can be confusing because one gene can generate several circular transcripts and one miRNA can become targets for different circRNAs. Luckily, existing databases, such as circBase (http://circbase.org/), allow one to assess the variety of names, forms and functions of circRNAs.

Functions
Although the exact regulatory mechanism of circRNAs is still unclear, current research concentrates mainly on the following four aspects, namely sponging of miRNAs, cis-regulating of parent genes, binding to proteins and encoding proteins (Figure 1). What is more, the categorization and localization of circRNAs are important to analyze their functions (Table 1).

CircRNAs as miRNA sponges
Lines of evidence indicate that natural circRNAs serve as effective miRNA sponges ( Figure 1D). MiRNAs are small noncoding RNAs (19-22 nt) which negatively modulate mRNA expression in post-transcriptional stage via binding to 3 -untranslated regions [29]. CircRNAs localized in cytoplasm can interact with both miRNA and AGO, and sequentially eliminate the suppression of miRNAs on mRNA. For example, the first seriously studied circRNA, Cdr1as, harbors more than 70 conserved miR-7 binding site. The effect of Cdr1as expression simulates that of miR-7 silencing, indicating a role of miRNA binding [10,11]. Another well-known circRNA, Sry, can serve as competitive inhibitor for miR-138 by binding to target sites [11]. Although one paper published in 2014 indicated that few circRNAs could act as efficient miRNA sponges [30], more and more circRNAs have been demonstrated in recent years to shape gene expression via inhibiting miRNAs. For example, circHIPK3 contains 18 potential binding sites for nine different miRNAs [31]. CircHIPK3 plays an important role in hepatocellular carcinoma (HCC) by sponging miR-124, leading to abnormal cancer cell proliferation and migration [32]. CircHIPK3 is also able to regulate endothelial proliferation and vascular dysfunction in diabetic retinopathy (DR) via blocking miR-30a-3p [33].

Cis-regulation of parent genes
CiRNAs are able to regulate the expression of their parent genes in cis ( Figure 1E). Nuclear ciRNAs localized near the transcription sites of their parent genes can interact with RNA Pol II elongation machinery and act as positive regulators for transcription. For example, knockdown of ci-ankrd52, ci-mcm5 and ci-sirt7 results in suppressed expression of their parental mRNAs [6]. EIciRNAs are also capable to cis-regulate their parental genes. EIciRNAs are able to promote transcription of RNA Pol II through interacting with U1 snRNP [34]. Silencing of two EIciRNAs, circEIF3J and circPAIP2, reduces the expression level of their parental mRNAs [5].
The majority of ecircRNAs acts as miRNA sponges and does not regulate the expression of their cognate genes. However, circ-ITCH shares the same miRNA response elements with the 3 -untranslated region of ITCH mRNA. By binding to miRNAs, circ-ITCH relieves the suppressive effect of miRNA on its own parent mRNA. The enhanced expression of ITCH at post-transcriptional level by circ-ITCH finally results in the suppressed activity of Wnt/β-catenin pathway. Dysregulation of circ-ITCH and Wnt/β-catenin pathway are involved in the progression of lung cancer, esophageal squamous cell carcinoma and colorectal cancer (CRC) [26,35,36].

Protein binding
Except miRNAs, circRNAs are also able to interact with other entities, such as proteins ( Figure 1F). In a study conducted by Du et al., circ-Foxo3 was demonstrated to influence cell cycle progression by binding to two cell cycle proteins (CDK2 and p21) [37]. PES1 is an important protein essential for ribosome biogenesis. CircANRIL can attach to PES1 and increase apoptosis of cells [38].
RNA binding proteins are a specific class of proteins which can regulate the formation and function of mRNAs through binding to ACUAA motifs in 3 UTR region [39]. HuR is a well-studied RNA binding protein which can adhere to PABPN1 mRNA and promote its expression. However, hsa circ 0031288 (CircPABPN1) is able to reduce PABPN1 expression via sequestering and suppressing HuR [27].

Protein coding
It was reported in 1995 that synthetic circRNAs with continuous open reading frames were capable to be translated into long-repeating polypeptide chains [40]. Several recent researches indicate that endogenous circRNAs containing open reading frames are also able to encode functional proteins ( Figure 1G) [41]. For example, FBXW7-185aa translated from circ-FBXW7 and SHPRH-146aa translated from circ-SHPRH are tumor suppressors in glioblastoma [28,42].

Nontumor diseases
In accordance with the disclosure of circRNA functions, the pivotal roles of circRNA in diseases as biomarker and regulator are catching more attention ( Table 2).

Cardiovascular diseases Atherosclerosis
Atherosclerosis is a disease in which the inside of an artery narrows due to building up of plaques, and may lead to coronary artery disease (CAD), stroke and peripheral artery disease. In a 2016 research conducted by Holdt et al., the atheroprotective function of circANRIL was investigated. Forced expression of circANRIL could increase cell apoptosis and decrease proliferation in vitro [38]. Instead of sponging miRNAs [11] or cis-regulating parental genes [6], circANRIL implemented its protective function through binding to PES1 protein, which finally resulted in the impaired ribosome maturation as well as apoptosis in smooth muscle cells and macrophages [38].
In a more recent study, Li et al. constructed an oxLDL-treated endothelial cell injury model in order to find potential diagnostic biomarker and elucidate molecular mechanism for atherosclerosis. They found that hsa circ 0003575 was significantly upregulated in oxLDL-treated endothelial cells. Hsa circ 0003575 silencing could promote cell proliferation and angiogenesis in ox-LDL-treated endothelial cells [43].

Coronary artery disease
CAD, also named coronary atherosclerotic heart disease, is caused by coronary artery atherosclerotic stenosis or occlusion and the sequential myocardial ischemia and hypoxia. In a current study, Pan et al. investigated circRNAs in three pairs of plasma samples from CAD patients and control subjects. 24 different circRNAs were identified, and in the network constructed by bioinformatic technology, nine circRNAs could together promote TRPM3 expression by inhibiting hsa-miR-130a-3p [19]. Although this research provides a new insight into pathological regulation of CAD, the interactions in this network are mainly based on bioinformatics analysis, and demand further investigation.
CAD is closely related with Type 2 diabetes mellitus (T2DM), and the metabolism abnormalities in T2DM directly increase CAD risk [54]. By using microarray analysis, Li et al. explored circRNA expression profile in CAD patients with hyperglycemia. Logistic regression analysis of two independent cohorts showed that hsa-circ11783-2 was more correlated with both CAD and T2DM than other selected circRNAs [44]. This is the first study to examine the association of circRNAs with both CAD and T2DM, and the function of hsa-circRNA11783-2 requires further exploration.

Myocardial infarction & heart failure
Most myocardial infarction (MI) occurs on the bases of CAD, when blood supply suddenly decreases or stops to a part of the heart. Severe MI causes cardiomyocyte death and may provoke heart failure.
In a study concerning the pathological mechanisms of MI, the authors discovered that circRNA MFACR regulated cardiomyocyte death via circMFACR/miR-652-3p/MTP18 axis [45]. MTP18 was able to induce mitochondrial fission [55] and promote cardiomyocyte apoptosis in MI [45]. MiR-652-3p was capable of inhibiting MTP18 expression and was negatively regulated by circMFACR. CircRNA MFACR elevated apoptotic cell death in MI through eliminating miR-652-3p suppression on MTP18. This research sheds new light on understanding molecular mechanism of MI [45].
A large percentage of acute MI (AMI) patients develop into heart failure due to maladaptive left ventricular remodeling. Salgado-Somoza et al. identified circRNA MICRA as a prognostic biomarker to improve the risk stratification after AMI. Patients with decreased MICRA level were more likely to be classified into reduced ejection fraction group. Both ordinal regression analysis and bootstrap internal validation were utilized to demonstrate the value of MICRA in prognostic stratification of AMI induced heart failure [46].

Hypertension
In a research published in 2016, Wu et al. identified circRNA expression profiles in hypertension patients' peripheral blood. Hsa circ 0005870 showed a significant down expression in hypertension group. Both GO and KEGG pathway analysis indicated the involvement of hsa circ 0005870 in hypertension [47]. Hsa circ 0005870 may represent a novel biomarker for the diagnosis of hypertension. However, the exact regulation mechanism of hsa circ 0005870 needs further investigation.
In another study, Cheng et al. investigated characteristic profile of circRNA in kidney samples from four kinds of hypertension rat models. Aberrant circRNAs were identified and verified with RT-qPCR. Bioinformatics technologies were used to predict the circRNA/miRNA/mRNA network [56]. This study serves as a primary foundation for further researches concerning hypertension combined with kidney diseases.

Cardiomyopathy
As a kind of cardiomyopathy, the pathological progression of myocardial fibrosis is characterized with the activation of cardiac fibroblasts (CFs), in which fibroblasts transform into myofibroblasts, resulting in collagen deposition in extracellular matrix [57]. Myocardial fibrosis can be caused by deformity of cardiomyocytes resulting from metabolism disorders in diabetes mellitus patients. In two recent researches concerning circRNAs in myocardial fibrosis, both diabetic mouse and mice CFs were used as research models. CircRNA circRNA 000203 [48] and circRNA 010567 [49] were found to be remarkably upregulated in diabetic mouse myocardium and Ang-II-treated CFs. Elevated level of circRNA 000203 and circRNA 010567 accelerated fibrosis-associated protein expression. For mechanism, circRNA 000203 was demonstrated to inhibit miR-26b-5p and eliminate miRNA suppressive effect on Col1a2 and CTGF [48]. Another circRNA, circRNA 010567, was proved to sponge miR-141 and increase the expression level of TGF-β1. [49]. These two studies shed new light on the pro-fibrosis effect of circRNA 000203 and circRNA 010567, and the regulatory function of circRNA/miRNA/mRNA axis in myocardial fibrosis.
Cardiac hypertrophy is characterized by maladaptive thickening of the myocardium. Wang et al. revealed the circHRCR/miR-223/ARC regulatory axis in cardiac hypertrophy, which might finally develop into abnormal cardiac remodeling and heart failure. MiR-223 was capable to induce heart failure in vivo and cardiomyocyte hypertrophy in vitro [50]. ARC, a reported protein involved in pathological hypertrophy inhibition [58], was demonstrated to be the downstream target of miR-223. In order to find the antihypertrophy molecule, the authors selected 100 published circRNAs from online databases. Among them, circHRCR was significantly downregulated in pathological conditions. Further investigation demonstrated that circHRCR could repress abnormal cardiac hypertrophy and heart failure through the circHRCR/miR-223/ARC axis [50].

Diabetes mellitus Diabetes mellitus
Diabetes mellitus is a kind of metabolic disorder in which patients are affected by hyperglycemia due to inadequate insulin or insulin resistance. Diabetes mellitus can be divided into Type 1 diabetes mellitus and T2DM.
In 2016, Zhao et al. delineated the expression profile of circRNAs in T2DM and prediabetes patients' peripheral blood for the first time. They selected five circRNAs as candidate biomarkers and verified in two independent cohorts. The results showed that hsa circ 0054633 presented the highest diagnostic ability among the chosen circRNAs [13]. This study provides an insight into novel biomarker for prediabetes and T2DM.
Long-term diabetes mellitus often leads to vascular complications, including microvascular and macrovascular diseases, which are the major causes for morbidity and mortality in diabetes mellitus. CircRNAs are involved in diabetes mellitus correlated vasculopathy. For example, circWDR77 is upregulated in high glucose treated vascular smooth muscle cells. CircWDR77 regulates vascular smooth muscle cells proliferation and migration via directly binding to miR-124 and alleviating suppression for target FGF-2 [59].

Diabetic retinopathy
DR is one of the common complications caused by diabetes mellitus. In 2017, Gu et al. analyzed the altered circRNA profiles in DR patients' serum. This is the first circRNA study concerning DR and lays the first stone for later biomarker detection and mechanism elucidation [60]. In a following study, Zhang et al. revealed the distinctive expression profile of circRNAs in diabetic retinas. Circ 0005015 expression was significantly upregulated in retina samples, vitreous samples, peripheral plasma samples and fibrovascular membranes of DR patients. Circ 0005015 silencing reduced human retinal vascular endothelial cells (HRVECs) proliferation, migration and tube formation. Luciferase activity assays found miR-519d-3p as the direct target for circ 0005015 [51]. Circ 0005015 is manifested as a regulatory biomarker for the DR diagnosis and treatment. In another research, Shan et al. demonstrated that circHIPK3 was elevated in hyperglycemia treated retinal endothelial cells and diabetic mouse retinas. CircHIPK3 served as a miRNA sponge to block miR-30a-3p activity and thus induced increase in levels of VEGFC, WNT2 and FZD4 [33]. Increment of VEGFC, WNT2 and FZD4 was reported in various retina disorders [61,62]. CircHIPK3 silencing could alleviate diabetes-induced endothelial proliferation and retina microvascular dysfunction [33].

Regulate insulin secretion
Several reports manifest that circRNAs are involved in the regulation of islet cells vitality. For example, Cdr1as, perhaps the best-identified endogenous mammalian circRNA, can be increased in islet cells by long-term forskolin and PMA stimulation [63]. As an miR-7 sponge [30], Cdr1as is able to improve insulin secretion and transcription through inhibiting miR-7 and accelerating Myrip and Pax6 expression [63]. In another study, the authors analyzed circRNAs in human islets and cognate ones in mouse islets. They revealed that Cdr1as and circHIPK3 were abundant in normal islets, but declined in diabetic mouse. Cdr1as and circHIPK3 silencing in wild-type animal models caused defective insulin secretion and diminished islet cell proliferation. While Cdr1as performed such regulatory function by blocking miR-7, circHIPK3 regulated islet cell function by sequestering miR-124-3p and miR-338-3p and elevating Slc2a2, Akt1 and Mtpn [64].

Rheumatoid arthritis
Zheng et al. [65] and Ouyang et al. [52] screened the expression profile of circRNAs in rheumatoid arthritis (RA) patients' peripheral mononuclear cells by microarray analysis. The assay results were verified by RT-qPCR method. Zheng et al. predicted the circRNA/miRNA interaction utilizing bioinformatic software [65]. Ouyang et al. analyzed the correlation between differential circRNAs and clinicopathological factors, finding that circRNA 104871 exhibited the largest diagnostic ability [52].

System lupus erythematosus
Li et al. screened circRNA profiles in system lupus erythematosus patients' peripheral blood plasma. CircRNA candidates were selected and validated. Potential circRNA/miRNA interaction networks were constructed [66]. In another study, Luan et al. determined circRNA profiles in renal samples from lupus nephritis patients and health controls [53]. In their preceding investigation, the authors noticed that miR-150 was positively correlated with renal chronicity scores [67]. In the current study, they spotted circHLA-C as a probable regulator for miR-150. CircHLA-C and miR-150 exhibited a negative correlation. As a potential biomarker, circHLA-C was positively correlated with clinical factors, such as serum creatinine, renal activity index, proteinuria and crescentic glomeruli [53].

Cancers
As it has been revealed by numerous studies, circRNAs are involved in the initiation and progression of various human cancers and may become potential diagnostic biomarkers, as it is shown in Table 3.

Lung cancer
Lung cancer, also known as lung carcinoma, is the leading cause of cancer death globally [103]. According to histopathological classification, lung cancer can generally be divided into non-small-cell lung cancer (NSCLC) and small cell lung cancer.
Luo et al. reported that hsa circ 0000064 exhibited elevated expression in both lung cancer tissues and lung cancer cell lines (A549 and H1229). There was a close correlation between hsa circ 0000064 augmentation and tumor differentiation, tumor-lymph node-metastasis (TNM) stage and lymphatic metastasis. Hsa circ 0000064  silencing suppressed the proliferation and migration of cancer cells, and promoted cell apoptosis. For molecular mechanisms, the authors discovered that hsa circ 0000064 regulated apoptotic-related proteins, cycle-related proteins and invasion-related proteins [68]. NSCLC accounts for nearly 85% of all primary lung cancers. Yao et al. revealed that circRNA 100876 was elevated in NSCLC tissues compared with pair-matched nontumor tissues. The upregulated expression was correlated with TNM stage and lymphatic metastasis. Furthermore, NSCLC patients with higher circRNA 100876 level had an overall shorter survival time than NSCLC patients with lower expression level [69]. In another study, hsa circ 0014130 was found to be notably upregulated in NSCLC tissues and the expression was also associated with both TNM stage and lymphatic metastasis. The AUC was 0.878. Then, the circRNA/miRNA interaction network was predicted by bioinformatics technologies [70].
Lung adenocarcinoma (LAC) is currently the most common subtype of NSCLC in lifelong nonsmokers [104]. Zhao et al. investigated circRNA profile in early-stage LAC patients' tumor tissue and adjacent normal tissue. Five dysregulated circRNAs were validated and potential circRNA/miRNA network was predicted [105]. In another study, hsa circ 0013958 was detected as a LAC biomarker with regulatory potency. Hsa circ 0013958 was increased in LAC cell lines, cancer tissues and cancer patients' plasma. The expression level of hsa circ 0013958 was associated with tumor staging and lymph node metastasis and the AUC was 0.815. Biological function experiments validated that hsa circ 0013958 was involved in cellular proliferation and metastasis. Finally, the hsa circ 0013958/miR-134/CCND1 regulatory axis was constructed [71]. Similarly, the expression level of hsa circ 0012673 was significant overexpressed in LAC tissues, and was associated with tumor size. Hsa circ 0012673 could accelerate LAC proliferation through sequestering miR-22, resulting in elevated level of targeted ErbB3 [72].

Breast cancer
The most commonly diagnosed cancer among women has been breast cancer in recent years, which is also the leading cause of cancer death in women younger than 45 years [106]. In a present study, a circRNA with antioncogenic role was found in breast cancer. CircRNA-000911 was significantly decreased in breast cancer tissues and cancer cell lines. Overexpression of circRNA-000911 could induce decreased cell proliferation, migration and invision as well as increased apoptosis. CircRNA-000911 implemented its regulatory function via inhibiting miR-449a and promoting Notch1 and NF-κB signaling pathway [73]. In the study performed by Liang et al., the authors revealed that circ-ABCB10 was significantly increased in breast cancer tissues. SiRNA induced circ-ABCB10 silencing contributed to decreased cancer cell proliferation and elevated apoptosis. Function analysis demonstrated that circ-ABCB10 carried out its tumorigenesis role via binding to miR-1271 [74]. Similar tumor-promoting effect was also observed in hsa circ 0001982 targeting miR-143 [75].
Triple negative breast cancer is a kind of breast cancer not sensitive to hormone therapies targeting ER, PR or Her2/neu. He et al. analyzed circRNA patterns in cancer cell lines and verified one upregulated circRNA circGFRA1. Silencing of circGFRA1 relieved suppression for target miR-34a, which results in inhibited cancer cell proliferation and promoted apoptosis [76]. Yan et al. revealed that circVRK1 was reduced in breast cancer stem cells. CircVRK1 repressed stemness-maintenance ability of breast cancer stem cells [107]. Furthermore, miR-153-5p, the predicted target of circVRK1, was previously reported to be involved in stemness-maintenance of triple negative breast cancer [108].

Gastric cancer
Gastric cancer (GC) is among the leading causes of cancer death in developing countries nowadays [109]. Several researches delineated the global expression patterns of circRNAs in GC tissues or patients' plasma [79,82,110,111]. Part of the results was validated with RT-qPCR. These investigations lay the foundation for future exploration in GC.
From 2015 to 2017, quite a few circRNAs were detected as potential GC biomarkers, such as hsa circ 0074362 [77], hsa circ 002059 [78] and hsa circ 0014717 [79]. All of these diagnostic biomarkers were dysregulated in cancer tissue compared with paired noncancerous tissue and were correlated with several clinicalpathological factors. For example, Hsa circ 0074362 was significantly down-expressed in GC tissues, cancer cell lines and gastritis. The level of hsa circ 0074362 in GC was significantly lower compared with moderate gastritis. The expression level of hsa circ 0074362 was associated with CA19-9 and lymphatic metastasis. The receiver operator curve was 0.630 [77]. These suggest that hsa circ 0074362 may have potential values in the screening of GC.
Some circRNAs as biomarkers not only exist in tumor tissues, but also exist in patients' plasma, such as hsa circ 0000190 [80], hsa circ 0000520 [81], hsa circ 0001017 and hsa circ 0061276 [82]. For instance, Li et al. demonstrated that hsa circ 0001017 and hsa circ 0061276 expression levels were downregulated in both cancer tissues and paired plasma and were associated with tumor size and distal metastasis. Through combining the expression levels of hsa circ 0001017 and hsa circ 0061276 in tissue and plasma, the AUC could ascend to 0.966 [82].
Among these disclosed circular biomarkers, a few are manifested as functional miRNA sponges. In their previous study, Zhang et al. noticed the remarkable decline of hsa circRNA 100269 in recurrent GC tissue [112]. In their follow-up examination, they manifested that hsa circRNA 100269 overexpression could prohibit tumor cell proliferation via absorbing miR-630 [83]. Another example of regulatory circRNAs in GC is circLARP4. Based on their preceding research concerning mRNA LATS1 as a tumor suppressor in GC [84], Zhang et al. explored the upstream regulatory network for LATS1. They revealed that circLARP4 could impede GC cell proliferation and invasion by targeting miR-424-5p, which led to raised LATS1. However, only early stage patients with a higher circLARP4 expression had a better outcome than patients with circLARP4 low expression. Similar trends could not be observed in late stage patients [113].
Hepatocellular carcinoma HCC is the most common type of primary liver cancer and the third leading cause of cancer mortality in many countries [114]. Early researches demonstrated diagnostic potency of circRNAs in HCC. For instance, both hsa circ 0001649 and hsa circ 0005075 were identified as novel biomarkers for HCC [85,86]. Later researches concern not only about diagnostic ability but also regulatory potency of circRNAs in HCC. Some circRNAs display antioncogenic effects. Yao discovered that, in HCC tissue samples, both circZKSCAN1 and ZKSCAN1 mRNA were significantly lower compared with noncancer tissues. Blocking circZKSCAN1 and/or ZKSCAN1 mRNA would promote cancer cell proliferation, migration and invasion. Furthermore, circZKSCAN1 was associated with tumor numbers, cirrhosis, vascular invasion and tumor grade [87]. Another two circRNAs, circMTO1 and cSMARCA5, were also downregulated in HCC tissue. Both of them could inhibit HCC cell proliferation and migration. CircMTO1 exerted tumor suppressive role via circMTO1/miR-9/p21 axis [88], and cSMARCA5 via cSMARCA5/miR-17-3p, miR-181b-5p/TIMP3 axis [89]. Some circRNAs act as tumor-promoting molecules in HCC. Circ 0067934 and circHIPK3 were all highly expressed in HCC tissues. Knockdown of these circRNAs resulted in suppressed HCC cell proliferation and migration [32,90]. Circ 0067934 achieved its regulatory function through circ 0067934/miR-1324/FZD5 pathway [90], and circHIPK3 through circHIPK3/miR-124/AQP3 axis [32].
A majority of HCC in Asian area arises from chronic hepatocellular B virus (HBV) infection and subsequent cirrhosis [115]. Cui et al. clarified circRNA expression profiles in HBV-related HCC by microarray analysis [116]. Huang et al. discovered that circRNA 100338 was upregulated in HBV-related HCC, and correlated with both low survival rate of patients and invasive process of cancer cells. CircRNA 100338 could facilitate cancer cell invasion and migration by sponging and inhibiting miR-141-3p [91].

Colorectal cancer
In recent years, several circRNAs are detected as candidate biomarkers for CRC. For instance, the expression level of circRNA0003906 was significantly decreased in CRC tissues and CRC cell lines, and was correlated with clinicopathological factors, such as lymphatic metastasis and poor differentiation. The AUC was 0.818, which demonstrated the diagnostic ability of circRNA0003906 [92]. Similarly, hsa circ 001988 [93], hsa circRNA 103809 and hsa circRNA 104700 [94] were also demonstrated as potential diagnostic biomarkers. Their expression levels were dysregulated in CRC and correlated with clinicopathological features. The AUC for hsa circ 001988, hsa circRNA 103809 and hsa circ 104700 was 0.788, 0.699 and 0.616, respectively [93,94].
Several ectopic circRNAs were verified to have regulatory potential in CRC. Gua et al. revealed that hsa circ 0000069 silencing notably attenuated tumor cell proliferation, migration and invasion [95]. Zhang et al. testified that hsa circ 0007534 silencing led to suppressed proliferation and induced apoptosis of CRC cells [96]. In another research, the authors delineated that hsa circ 0020397 performed its tumor-generating effects via hsa circ 0020397/miR-138/TERT, PD-L1 axis [97].
Metastasis is among the major causes of tumor death and often occurs in late stages of tumor development [117]. Two recent studies investigated the metastasis in CRC by utilizing circRNA profiling. Jiang et al. detected differential circRNAs in CRC metastasis cells [118], and Zeng et al. compared circRNA expression in CRC patients with and without lung metastasis [119]. These findings provided candidate circRNAs for later investigations concerning CRC metastasis.

Bladder cancer
Recent studies reveal certain circRNAs with regulatory function in bladder cancer. Zhong et al. discovered that forced expression of circTCF25 would sequester miR-103a-3p and miR-107, which led to the increased CDK6 expression and promoted cancer cell proliferation and migration [98]. In another study, the authors manifested that circRNA-MYLK and its downstream target miR-29a/VEGFA, VEGFR2 signaling pathway were related with bladder cancer cell proliferation and epithelial-mesenchymal transition process [99]. Li et al. discovered that increased expression of circRNA BCRC4 could induce bladder cancer cell apoptosis by inhibiting miR-101 ability, which relieved suppression for EZH2 [100].
Circ-ITCH is generated from itchy E3 ubiquitin protein ligase (ITCH) coding region and is involved in bladder cancer. For instance, Yang et al. found that circ-ITCH was downregulated in bladder cancer tissues and cell lines. Decreased expression of circ-ITCH was correlated with shortened survival in bladder cancer patients. Expression of circ-ITCH represses cancer cell proliferation, migration and metastasis. This consequence of circ-ITCH was achieved by inhibiting miR-17 and miR-224 and promoting p21 and PTEN [101].

Glioma
In 2016, Song et al. constructed a computational filter named UROBORUS and disclosed circRNA expression in gliomas for the first time [120]. Then Li et al. revealed that hsa circ 0046701 was increased in glioma tissue and cell lines. Hsa circ 0046701 was able to regulate tumor cell proliferation and invasion by sponging miR-142-3p and increasing ITGB8 [102]. Interestingly, the open reading frame equipped in circ-FBXW7 and circ-SHPRH allowed them to translate functional proteins, FBXW7-185aa and SHPRH-146aa [28,42]. Both FBXW7-185aa and SHPRH-146aa displayed antioncogenic potency in glioma. These two researches indicated a potential new role for circRNAs.

Conclusion & future perspective
Unlike linear RNAs, continuous circRNAs have no free ends and are more resistant to RNase. The properties of circRNAs include stability, prevalence, specificity and conservation. CircRNAs have been demonstrated to regulate cellular function through sponging other factors (miRNAs or proteins), promoting transcription or coding proteins. From cancers to noncancerous disorders, large quantities of studies have revealed the involvement of circRNAs in human diseases as biomarkers and/or regulators. These potential circular biomarkers/regulators, combined with currently widely used diagnostic and treating methods, may improve future clinical activities. However, in order to contribute to later diagnosis and treatment, the precise role of circRNA in both physiological and pathological conditions needs further investigation. Besides the reported mechanism, new hypothesis may be proposed and demonstrated. What is more, the curative effects and the side effects of circRNAs as treatment targets in vivo should be evaluated and examined.

Executive summary
• Circular RNAs (circRNAs) are generated without 3 and 5 free ends.
• CircRNAs regulate gene expression and transcription through binding to miRNAs, Pol II and proteins. Some endogenous circRNAs code proteins themselves.
• CircRNAs are related with diverse human diseases.
Authors' contributions Y Fang collected information and wrote the manuscript.

Financial & competing interests disclosure
The authors have no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.
No writing assistance was utilized in the production of this manuscript.

Open access
This work is licensed under the Creative Commons Attribution 4.0 License. To view a copy of this license, visit http://creativecomm ons.org/licenses/by/4.0/